CN112348170A - Fault diagnosis method and system for turnout switch machine - Google Patents
Fault diagnosis method and system for turnout switch machine Download PDFInfo
- Publication number
- CN112348170A CN112348170A CN202011248062.6A CN202011248062A CN112348170A CN 112348170 A CN112348170 A CN 112348170A CN 202011248062 A CN202011248062 A CN 202011248062A CN 112348170 A CN112348170 A CN 112348170A
- Authority
- CN
- China
- Prior art keywords
- current curve
- curve data
- sample set
- fault
- network model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 67
- 238000003745 diagnosis Methods 0.000 title claims abstract description 44
- 238000004458 analytical method Methods 0.000 claims abstract description 71
- 238000005070 sampling Methods 0.000 claims abstract description 55
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 18
- 238000012549 training Methods 0.000 claims description 71
- 239000010410 layer Substances 0.000 claims description 51
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 26
- 238000004590 computer program Methods 0.000 claims description 12
- 230000002159 abnormal effect Effects 0.000 claims description 10
- 230000007257 malfunction Effects 0.000 claims description 10
- 230000002708 enhancing effect Effects 0.000 claims description 9
- 238000013524 data verification Methods 0.000 claims description 8
- 239000011229 interlayer Substances 0.000 claims description 8
- 238000007781 pre-processing Methods 0.000 claims description 8
- 238000011176 pooling Methods 0.000 claims description 7
- 230000003287 optical effect Effects 0.000 claims description 6
- 238000007405 data analysis Methods 0.000 claims description 5
- 230000006870 function Effects 0.000 claims description 4
- 230000003993 interaction Effects 0.000 abstract description 4
- 238000010606 normalization Methods 0.000 abstract description 4
- 238000010586 diagram Methods 0.000 description 11
- 230000008569 process Effects 0.000 description 8
- 238000012545 processing Methods 0.000 description 8
- 238000013528 artificial neural network Methods 0.000 description 6
- 238000012360 testing method Methods 0.000 description 6
- 238000004891 communication Methods 0.000 description 5
- 238000012935 Averaging Methods 0.000 description 3
- 238000012795 verification Methods 0.000 description 3
- 230000007547 defect Effects 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 208000027418 Wounds and injury Diseases 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000006378 damage Effects 0.000 description 1
- 238000013502 data validation Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 208000014674 injury Diseases 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000009897 systematic effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The embodiment of the invention provides a method and a system for diagnosing faults of a switch machine, which comprises the following steps: acquiring a current curve of a turnout switch machine; sampling the current curve based on a preset length to obtain current curve data; inputting the current curve data into a fault analysis network model so as to obtain a fault diagnosis result corresponding to the current curve according to an output result of the fault analysis network model; wherein the fault analysis network model is a deep convolutional neural network with a residual error structure. According to the turnout switch machine fault diagnosis method and system provided by the embodiment of the invention, the fault reason corresponding to the fault category with the highest probability is obtained by constructing the deep convolutional neural network with the residual error structure, so that the interaction information between different phases of the current curve can be better extracted, and the network structure can be as deep as possible due to the residual error structure and the hierarchy normalization, so that the difference information between normal and fault samples can be better learned, and the prediction precision is improved.
Description
Technical Field
The invention relates to the technical field of rail transit, in particular to a method and a system for diagnosing faults of a turnout point switch.
Background
The switch machine is a switch device of the switch, and can be used for switching the switch or locking the switch. Failure of a switch machine may result in derailment of an on-track train, resulting in significant economic loss and loss of life and personal injury. The current time sequence of the turnout switch machine can be used for judging the working state of the turnout switch machine, such as whether the turnout switch machine is in a normal working state or not, and if the turnout switch machine is in an abnormal working state, the current time sequence is used for judging which abnormal working state the turnout switch machine is in.
The core of judging the turnout working state according to the turnout current time sequence is to judge whether the current turnout current curve is a normal or abnormal current curve based on historical turnout current curve data. The current time sequence fault diagnosis models of the turnout switch machine are mainly divided into two types:
the first is a matching-based fault diagnosis method. The method carries out fault diagnosis by comparing the similarity between a current sample to be evaluated and a fault sample in a historical database. The method comprises the steps of firstly collecting a large number of fault sample examples, then calculating the appointed distance measurement between a sample to be evaluated and each sample in a fault sample database, wherein the appointed distance measurement is Dynamic Time Warping (DTW) generally, so as to find the most similar fault curve and judge the possible fault type of the most similar fault curve. The method cannot abstract the sample characteristics of the fault, and cannot perform correct fault diagnosis when the sample which is not in the fault database is faced.
The second is a fault diagnosis method based on statistical machine learning and deep learning. The method trains a classifier by extracting deep features of a fault sample, and then classifies the sample to be evaluated. The method needs to extract the relevant characteristics of the fault sample and the normal sample for learning. However, the method simply depends on the deep neural network, and the local characteristics of the time series data cannot be effectively acquired, so that the prediction accuracy is poor.
In view of the above, there is a need to improve the existing fault diagnosis method of the point switch machine to improve the accuracy and robustness of fault prediction.
Disclosure of Invention
The embodiment of the invention provides a method and a system for diagnosing faults of a turnout switch machine, which are used for solving the defect of poor prediction precision caused by the fact that local characteristics of time sequence data cannot be effectively acquired when the faults of the turnout switch machine are diagnosed in the prior art.
In a first aspect, an embodiment of the present invention provides a method for diagnosing a fault of a switch machine, which mainly includes:
acquiring a current curve of a turnout switch machine; sampling the current curve based on a preset length to obtain current curve data; inputting the current curve data into a fault analysis network model so as to obtain a fault diagnosis result corresponding to the current curve according to an output result of the fault analysis network model; wherein the fault analysis network model is a deep convolutional neural network with a residual error structure.
Optionally, the sampling the current curve based on the preset length to obtain current curve data mainly includes: under the condition that the length of the current curve is larger than the preset length, down-sampling the current curve to acquire current curve data; and under the condition that the length of the current curve is smaller than the preset length, performing up-sampling interpolation on the current curve to obtain the current curve data.
Optionally, before inputting the current curve data into the fault analysis network model, the method may further include:
acquiring a normal current curve sample and an abnormal current curve sample, and constructing a current curve sample set; sampling each sample in the current curve sample set to obtain a current curve data sample set; carrying out layered sampling on the current curve data sample set to obtain a current curve data training sample set; enhancing the current curve data training sample set based on a sliding window method to obtain an enhanced current curve data training sample set; combining the enhanced current curve data training sample set and the current curve data training sample set to form a model training sample set; and pre-training the fault analysis network model by using the model training sample set and the diagnosis labels corresponding to the samples in the model training sample set.
Optionally, the enhancing the current curve data training sample set based on the sliding window method to obtain an enhanced current curve data training sample set mainly includes:
determining the size and the sliding step length of a sliding window; sampling any current curve data in the current curve data training sample set based on the size of the sliding window and the sliding step length to obtain a plurality of enhanced samples to construct a local sample set; the length of each of the enhancement samples in the local sample set is the size of the sliding window; and (3) after each enhancement sample is supplemented with 0 backwards until the length of the enhancement sample is the preset length, acquiring the enhancement current curve data training sample set.
Optionally, after the pre-training of the fault analysis network model, the method further includes: verifying the pre-trained fault analysis network model by using a current curve data verification set; the current curve data verification set is obtained after the current curve data sample set is subjected to layered sampling.
Optionally, the structure of the fault analysis network model specifically includes: the system comprises a total input layer, a first one-dimensional residual block, a maximum pooling layer, a one-dimensional Spatial Dropout layer, a second one-dimensional residual block, a global average pooling layer, a Softmax function layer and a total output layer which are connected in sequence; the first one-dimensional residual block and the second one-dimensional residual block have the same structure.
Optionally, the structures of the first one-dimensional residual block and the second one-dimensional residual block specifically include: the multilayer optical amplifier comprises an interlayer input Layer, a first one-dimensional Layer Normailization Layer, a second one-dimensional Layer Normailization Layer, a third one-dimensional Layer Normailization Layer, an Add Layer Normailization Layer and an interlayer output Layer which are sequentially connected.
In a second aspect, an embodiment of the present invention further provides a system for diagnosing a fault of a switch machine, which mainly includes a data acquisition unit, a data preprocessing unit, and a data analysis unit, wherein:
the data acquisition unit is used for acquiring a current curve of the turnout switch machine;
the data preprocessing unit is used for sampling the current curve based on a preset length to obtain current curve data;
the data analysis unit is used for inputting the current curve data into a fault analysis network model so as to obtain a fault diagnosis result corresponding to the current curve according to an output result of the fault analysis network model;
wherein the fault analysis network model is a deep convolutional neural network with a residual error structure.
In a third aspect, an embodiment of the present invention further provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the steps of the method for diagnosing a fault of a switch machine as described in any one of the above.
In a fourth aspect, the present invention further provides a non-transitory computer readable storage medium, on which a computer program is stored, wherein the computer program, when executed by a processor, implements the steps of the method for diagnosing the fault of the point switch machine as described in any one of the above.
According to the turnout switch machine fault diagnosis method and system provided by the embodiment of the invention, the fault reason corresponding to the fault category with the highest probability is obtained by constructing the deep convolutional neural network with the residual error structure, so that the interaction information between different phases of the current curve can be better extracted, and the network structure can be as deep as possible due to the residual error structure and the hierarchy normalization, so that the difference information between normal and fault samples can be better learned, and the prediction precision is improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and those skilled in the art can also obtain other drawings according to the drawings without creative efforts.
Fig. 1 is a schematic flow chart of a method for diagnosing a switch point machine fault according to an embodiment of the present invention;
fig. 2 is a timing diagram of a three-phase current curve of a switch point switch according to an embodiment of the present invention;
FIG. 3 is a schematic flow chart illustrating a process for enhancing a training sample set of current curve data according to an embodiment of the present invention;
fig. 4 is a schematic diagram of a data processing flow in a process of diagnosing a fault of a point switch machine according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of a fault analysis network model according to an embodiment of the present invention;
fig. 6 is a schematic structural diagram of a switch point machine fault diagnosis system according to an embodiment of the present invention;
fig. 7 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Fig. 1 is a schematic flow chart of a method for diagnosing a switch machine fault according to an embodiment of the present invention, as shown in fig. 1, the method includes, but is not limited to, the following steps:
step 11, obtaining a current curve of a turnout switch machine;
step 12, sampling the current curve based on a preset length to obtain current curve data;
wherein the fault analysis network model is a deep convolutional neural network with a residual error structure.
In the embodiment of the invention, a current reading device is additionally arranged in the control circuit of each turnout switch machine so as to record the current time sequence current of each turnout switch machine in real time during operation and draw a corresponding time sequence current curve. When the running state of the target turnout switch machine or the fault of the target turnout switch machine needs to be diagnosed, only a time sequence current curve of the target turnout switch machine in a target time period needs to be called, and the time sequence current curve is preprocessed, wherein the preprocessing comprises sampling by utilizing a preset length. And then, inputting the current curve data acquired after preprocessing into a fault analysis network model trained in advance so as to diagnose the fault.
According to the technical scheme, the deep convolutional neural network with the residual error structure is used as a fault analysis network model to analyze input current curve data and obtain a fault diagnosis result corresponding to the input current curve data, so that the input current curve data are required to be the same in length, and the fault analysis network model can accurately obtain the rule of sample distribution.
The deep convolutional neural network with the residual error structure is characterized in that the residual error structure is set in a common deep convolutional neural network model to realize that the difference between an actual value and a model estimation value (fitting value) is used as a residual error item, and the defects that the deep neural network is difficult to train, and gradient disappearance, gradient explosion and the like exist are overcome. The use of the residual block enables training of deeper neural networks. Therefore, the constructed fault analysis network model forms a deeper neural network model by accumulating at least one residual block together, and can acquire local characteristics of time sequence data more deeply on the basis of extracting deep characteristics of data without manually extracting the characteristics so as to be beneficial to improving the fault diagnosis precision of the point switch machine.
According to the turnout switch machine fault diagnosis method provided by the embodiment of the invention, the fault reason corresponding to the fault category with the highest probability is obtained by constructing the deep convolutional neural network of the residual error structure, so that the interaction information between different phases of the current curve can be better extracted, and the network structure can be as deep as possible due to the residual error structure and the hierarchy normalization, so that the difference information between normal and fault samples can be better learned, and the prediction precision is improved.
Based on the content of the foregoing embodiment, as an optional embodiment, the sampling the current curve based on the preset length to obtain current curve data includes:
under the condition that the length of the current curve is larger than the preset length, down-sampling the current curve to acquire current curve data;
and under the condition that the length of the current curve is smaller than the preset length, performing up-sampling interpolation on the current curve to obtain the current curve data.
Fig. 2 is a schematic diagram of a data processing flow in the process of diagnosing the fault of the point switch machine according to the embodiment of the present invention, as shown in fig. 2, the horizontal axis represents time step (time step) and the vertical axis represents current ampere (a). Fig. 2 shows the time series of the three-phase currents of the point switch machine (phaseA, phaseB and phaseC, respectively).
Specifically, the collected current curve of the point switch machine is a time sequence curve graph, and in the embodiment of the present invention, the time sequence curve graph is analyzed by a fault analysis network model to obtain a fault diagnosis result corresponding to the current curve. Therefore, the timing diagram needs to be preprocessed when being input into the fault analysis network model, so as to ensure that the data formats input into the fault analysis network model are the same.
Specifically, in the embodiment of the present invention, first sampling the collected and acquired timing graph includes: and (3) setting a parameter L with a preset length, and sampling each time sequence curve in the time sequence curves to the specified length L. For samples with length larger than L, down-sampling is carried out; and performing upsampling interpolation on samples with the length less than L.
Wherein, the down-sampling refers to: for the timing curve, the interval length L is sampled once, so that the resulting current curve data is a downsampling of the original timing curve.
Wherein the upsampling interpolation means: collecting current curve data, wherein the up-sampling is to convert continuous signals in time and amplitude into discrete signals in time and amplitude under the action of sampling pulse (with length of L),
in fact, both the upsampling and the downsampling are to resample the time sequence curve, by comparing the length of the current curve with the preset length L, the upsampling is called when the length is greater than the length L, the downsampling is called when the length is smaller than the length L, and the essence of the upsampling is to interpolate or interpolate.
Therefore, the embodiment of the invention can acquire the current curve data with consistent constants by sampling the time sequence curve chart of the target turnout switch machine so as to unify the data format input to the fault analysis network model, is convenient to analyze the turnout switch machine by using the fault analysis network model, and can effectively improve the identification capability of the system.
Based on the content of the foregoing embodiment, as an optional embodiment, before inputting the current curve data to the fault analysis network model, the method further includes:
acquiring a normal current curve sample and an abnormal current curve sample, and constructing a current curve sample set;
sampling each sample in the current curve sample set to obtain a current curve data sample set;
carrying out layered sampling on the current curve data sample set to obtain a current curve data training sample set;
enhancing the current curve data training sample set based on a sliding window method to obtain an enhanced current curve data training sample set;
combining the enhanced current curve data training sample set and the current curve data training sample set to form a model training sample set;
and pre-training the fault analysis network model by using the model training sample set and the diagnosis labels corresponding to the samples in the model training sample set.
Specifically, before analyzing the current curve data of the target switch machine by using the fault analysis network model in the embodiment of the present invention, the fault analysis network model needs to be pre-trained, which specifically includes, but is not limited to, the following steps:
firstly, historical current curve samples of a target turnout switch machine and other turnout switch machines of the same type are obtained, and a current curve sample set is constructed. Wherein the set of current profile samples includes normal current profile samples and abnormal current profile samples.
Further, based on the method adopted in the above embodiment, each historical current curve sample is sampled by using a preset length, and a current curve data sample corresponding to each historical current curve sample is obtained, so as to construct a current curve data sample set. And each current curve data sample is provided with a corresponding fault diagnosis label. For example: if the current curve data sample is a normal current curve sample, the fault diagnosis label is normal; and if the current curve data sample is an abnormal current curve sample, marking a specific fault type on the fault diagnosis label.
Further, the current curve data sample set constructed in the previous step is subjected to hierarchical sampling, so that a training set and a verification set are selected and obtained. It should be noted that in the hierarchical sampling process, consistency of the proportion of the labels of the fault samples in the training set and the test set needs to be maintained, so that consistency of distribution of the fault samples in the training set and the test set can be ensured, and a rule that a trained fault analysis network model can accurately obtain distribution of the samples can be ensured.
The hierarchical sampling refers to random sampling for each category, and is often a method adopted to ensure uniformity and representativeness in sampling space or type selection. The sampling in each layer may be simple random or systematic sampling, and random sampling may result in better samples if there is no special need.
Further, enhancement processing is carried out on the current curve data training sample set obtained in the last step based on a sliding window method, so as to obtain an enhanced current curve data training sample set.
Optionally, the enhancing the current curve data training sample set based on the sliding window method to obtain an enhanced current curve data training sample set mainly includes the following steps:
determining the size and the sliding step length of a sliding window;
sampling any current curve data in the current curve data training sample set based on the size of the sliding window and the sliding step length to obtain a plurality of enhanced samples to construct a local sample set;
the length of each of the enhancement samples in the local sample set is the size of the sliding window;
and (3) after each enhancement sample is supplemented with 0 backwards until the length of the enhancement sample is the preset length, acquiring the enhancement current curve data training sample set.
The sliding Window algorithm (Moving Window) controls the amount of traffic by limiting the maximum number of cells that can be received in each time Window. In the sliding window algorithm, the time window is slid forward every one cell time, and the length of the sliding is one cell time.
The data enhancement strategy adopted by the embodiment of the application mainly comprises the following contents:
firstly, presetting the size w of a sliding window and the size k of a step length, and training an ith curve T in a sample set for current curve datai∈RL×3In other words, a local sample in the sliding window is intercepted as an enhanced sample, the sliding window is moved forward according to the step size k, and the i-th curve T is traversediThen, the curve T for the ith curve can be obtainedi∈RL×3Local sample set of
Wherein h ═ [ (L-w)/k]+1, j is an intermediate variable. Wherein the length of each sample of the locally enhanced sample set of the ith sample is equal to the size w of the sliding window, and the local characteristic information of the sample (the ith curve) is contained.
Fig. 3 is a schematic flowchart of a process for enhancing any current curve data in a training sample set of current curve data according to an embodiment of the present invention, as shown in fig. 3, the parameters used are w-60, the step size k is 30, and the number of local samples obtained by each enhanced sample is 4. Accordingly, the fault diagnosis label of each of the enhanced samples is the same as the fault diagnosis label of the current curve data.
Further, each enhanced sample of the ith sample is subjected to backward 0 filling to reach the L length. The 0 complementing strategy facilitates the subsequent neural network to learn the local features during feature extraction.
According to the fault diagnosis method for the turnout switch machine, historical time sequence data are preprocessed, and each time sequence data is sampled to the same length; and then, a local sample of each time sequence is constructed by using a sliding window method, and the local characteristics of the sample are captured at a data level, so that the capture of the local characteristics of the data by the constructed fault analysis network model is enhanced by using a time sequence data enhancement strategy and a mode of enhancing a training data set, so that the fault can be better diagnosed by using the local characteristics of the model, the data enhancement is not performed, and the F1 value is obviously improved when a deep neural network is singly used for fault diagnosis.
Further, in the embodiment of the present invention, after the enhanced current curve data training sample set is obtained, the enhanced current curve data training sample set and the current curve data training sample set are combined to form a model training sample set, and the model training sample set is used to pre-train the fault analysis network model.
According to the embodiment of the invention, the model training sample set is formed by combining the enhanced current curve data training sample set and the current curve data training sample set which is not subjected to enhancement processing, so that the pre-training of the fault analysis network model is realized, the local characteristics of the sample are fully considered, the characteristics of the cause data set are not lost, the model training efficiency is effectively improved, and the robustness and the prediction precision of the model are improved.
Based on the content of the foregoing embodiment, as an optional embodiment, after the pre-training of the fault analysis network model, the method further includes: verifying the pre-trained fault analysis network model by using a current curve data verification set; the current curve data verification set is obtained after the current curve data sample set is subjected to layered sampling.
Fig. 4 is a schematic diagram of a data processing flow in a process of diagnosing a fault of a point switch machine according to an embodiment of the present invention, and as shown in fig. 4, in the embodiment of the present invention, a plurality of normal current curve data and a plurality of fault current curve data (i.e., abnormal current curve data) are first retrieved from a database of normal operating conditions of a point current and a database of abnormal operating conditions of the point current, respectively. Then, all current curve data are preprocessed, including sampling the current curve data through interpolation downsampling or upsampling by a preset length L, and the current curve data are filled to the preset length L to construct a current curve data sample set.
And further, interlayer sampling is carried out on the current curve data sample set based on interlayer sampling, and a current curve data training sample set and a current curve data verification set are respectively obtained. In the layer sampling process, the consistency of the proportion of the fault samples in the training set and the test set needs to be kept, and the method ensures the consistency of the distribution of the training set and the test set, thereby ensuring that the model can accurately obtain the rule of the distribution of the samples. Wherein, the ratio of the current curve data training sample set to the current curve data validation set can be set to 4 to 1.
Further, performing time series data enhancement on all samples in the current curve data training sample set, including:
under the condition of a given sliding window size and a sliding step length, each sample is subjected to sliding window sampling processing respectively to construct a local sample set.
Further, a local sample set after time sequence enhancement processing and current curve data which are not subjected to time sequence enhancement processing are used for training the sample set, and meanwhile, a pre-constructed fault analysis network model is pre-trained. And performing iterative verification on the fault analysis network model after each training by using a current curve data verification set which is not subjected to time sequence enhancement, and performing iterative selection and setting on model parameters according to a verification result.
According to the fault diagnosis method for the turnout switch machine, the consistency of the proportion of the fault sample labels in the training set and the testing set is kept in the layered sampling process, and the consistency of the distribution of the training set and the testing set is ensured, so that the fault analysis network model can accurately obtain the rule of sample distribution, and the prediction precision of the model can be improved.
Based on the content of the foregoing embodiment, as an optional embodiment, the structure of the fault analysis network model specifically includes: the system comprises a total input layer (Inputs), a first one-dimensional Residual Block (Conv1D Residual Block), a maximum Pooling layer (Maxbonding 1D), a one-dimensional Spatial Dropout layer, a second one-dimensional Residual Block (Conv1D Residual Block), a Global Averaging Pooling layer (Global Averaging Pooling1D), a Softmax function layer and a total output layer (Outputs) which are connected in sequence; the first one-dimensional residual block and the second one-dimensional residual block have the same structure.
Further, the structures of the first one-dimensional residual block and the second one-dimensional residual block specifically include: the multilayer optical amplifier comprises an interlayer input Layer, a first one-dimensional Layer Normailization Layer, a second one-dimensional Layer Normailization Layer, a third one-dimensional Layer Normailization Layer, an Add Layer Normailization Layer and an interlayer output Layer which are sequentially connected.
Fig. 5 is a schematic structural diagram of a fault analysis network model according to an embodiment of the present invention, and as shown in fig. 5, a network structure of the fault analysis network model is mainly formed by cascading modules of a Conv-1D Residual, and an overall framework of the fault analysis network model is shown on the right side of fig. 5, where a Residual Block (Conv1D Residual Block) structure is shown in a left side module of fig. 5.
And setting the preset length as L, and performing interpolation or truncation sampling on all current curve samples by utilizing the L. For a pre-created fault analysis network model, trained using a Batch Gradient Descent (BGD), the input of the fault analysis network model may be represented as (BatchSize, L, 3). In this representation, BatchSize represents the number of samples sent in this batch, and the values are usually 64 or 128; l is the length of the curve; the value 3 represents the dimension of the curve, which is 3 dimensions since it is a 3-phase current curve.
For the first one-dimensional Residual Block (Con1D Residual Block), the input is (BatchSize, L, 3), and since the structure of the Residual is adopted, and the same padding strategy is used, the total output size of the module is still (BatchSize, L, 3).
Further, after the sample of (BatchSize, L, 3) is input to the maximum pooling layer (MaxPooling1D), if the ranks of MaxPooling1D is set to 1 and pool _ size is set to 2, the output size is (BatchSize, L//2,3), where L//2 represents L divided by pool _ size quotient.
Further, after the sample of (BatchSize, L//2,3) passes through the one-dimensional Spatial Dropout module, some feature dimensions are set to 0. For example, for a sample, all values of phase a in phase 3 currents are set to 0, and only values of phase B and phase C are retained. The function of the one-dimensional Spatial Dropout module is to reduce the information redundancy of the current signals with different phases. The output latitude is the same as the input latitude, and the output size is still (BatchSize, L//2,3) after passing through the module.
Further, for Global Averaging Pooling1D, the module inputs the size (BatchSize, L//2,3) and outputs the size (BatchSize, 1, 3), and the module averages the features in the time dimension to obtain the final representation of the sample.
Further, for the Softmax module, the input size of the module is (BatchSize, 1, 3), and the output size is (BatchSize, 1, K), where K refers to the number of fault categories of the current fault diagnosis task. The module maps each sample in Batch onto a certain class label to achieve the purpose of classification.
According to the turnout switch machine fault diagnosis method provided by the embodiment of the invention, the deep convolution neural network with the residual error structure is designed for the fault diagnosis task, the fault of the turnout switch machine is diagnosed, the deep level features of data can be extracted without manually extracting the features, the adaptability to the time sequence task is good, the internal features of the time sequence can be extracted in a deep level, and the fault diagnosis precision is effectively improved.
Fig. 6 is a schematic structural diagram of a switch point machine fault diagnosis system according to an embodiment of the present invention, as shown in fig. 6, which mainly includes a data acquisition unit 1, a data preprocessing unit 2, and a data analysis unit 3, wherein:
the data acquisition unit is mainly used for acquiring a current curve of the turnout switch machine;
the data preprocessing unit is mainly used for sampling the current curve based on a preset length to obtain current curve data;
the data analysis unit is mainly used for inputting the current curve data into a fault analysis network model so as to obtain a fault diagnosis result corresponding to the current curve according to an output result of the fault analysis network model;
wherein the fault analysis network model is a deep convolutional neural network with a residual error structure.
Specifically, when the operation state of the target turnout switch machine or the fault of the target turnout switch machine needs to be diagnosed, only a time sequence current curve of the target turnout switch machine in a target time period needs to be called, and the time sequence current curve is preprocessed, wherein sampling is performed by using a preset length. And then, inputting the current curve data acquired after preprocessing into a fault analysis network model trained in advance so as to diagnose the fault.
According to the technical scheme, the deep convolutional neural network with the residual error structure is used as a fault analysis network model to analyze input current curve data and obtain a fault diagnosis result corresponding to the input current curve data, so that the input current curve data are required to be the same in length, and the prediction accuracy of the model is improved.
According to the turnout switch machine fault diagnosis system provided by the embodiment of the invention, the fault reason corresponding to the fault category with the highest probability is obtained by constructing the deep convolutional neural network of the residual error structure, so that the interaction information between different phases of the current curve can be better extracted, and the network structure can be as deep as possible due to the residual error structure and the hierarchy normalization, so that the difference information between normal and fault samples can be better learned, and the prediction precision is improved.
It should be noted that, when specifically executed, the system for diagnosing a fault of a point switch machine according to the embodiment of the present invention may be implemented based on the method for diagnosing a fault of a point switch machine according to any of the embodiments described above, and details of this embodiment are not described herein.
Fig. 7 illustrates a physical structure diagram of an electronic device, and as shown in fig. 7, the electronic device may include: a processor (processor)710, a communication interface (communication interface)720, a memory (memory)730 and a communication bus (bus)440, wherein the processor 710, the communication interface 720 and the memory 730 communicate with each other via the communication bus 740. The processor 710 may call the logic instructions in the memory 730 to execute the method for diagnosing the malfunction of the switch machine, which mainly comprises: acquiring a current curve of a turnout switch machine; sampling the current curve based on a preset length to obtain current curve data; inputting the current curve data into a fault analysis network model so as to obtain a fault diagnosis result corresponding to the current curve according to an output result of the fault analysis network model; wherein the fault analysis network model is a deep convolutional neural network with a residual error structure.
In addition, the logic instructions in the memory 730 can be implemented in the form of software functional units and stored in a computer readable storage medium when the software functional units are sold or used as independent products. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and the like.
In addition, the logic instructions in the memory 730 can be implemented in the form of software functional units and stored in a computer readable storage medium when the software functional units are sold or used as independent products. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
In another aspect, an embodiment of the present invention further provides a computer program product, where the computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program includes program instructions, and when the program instructions are executed by a computer, the computer is capable of executing the method for diagnosing a fault of a turnout switch provided by the above-mentioned embodiments of the method, and mainly includes: acquiring a current curve of a turnout switch machine; sampling the current curve based on a preset length to obtain current curve data; inputting the current curve data into a fault analysis network model so as to obtain a fault diagnosis result corresponding to the current curve according to an output result of the fault analysis network model; wherein the fault analysis network model is a deep convolutional neural network with a residual error structure.
In another aspect, an embodiment of the present invention further provides a non-transitory computer-readable storage medium, on which a computer program is stored, where the computer program is implemented by a processor to execute the method for diagnosing a fault of a switch machine provided in the foregoing embodiments, and the method mainly includes: acquiring a current curve of a turnout switch machine; sampling the current curve based on a preset length to obtain current curve data; inputting the current curve data into a fault analysis network model so as to obtain a fault diagnosis result corresponding to the current curve according to an output result of the fault analysis network model; wherein the fault analysis network model is a deep convolutional neural network with a residual error structure.
The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.
Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (10)
1. A method of diagnosing a malfunction of a switch machine, comprising:
acquiring a current curve of a turnout switch machine;
sampling the current curve based on a preset length to obtain current curve data;
inputting the current curve data into a fault analysis network model so as to obtain a fault diagnosis result corresponding to the current curve according to an output result of the fault analysis network model;
wherein the fault analysis network model is a deep convolutional neural network with a residual error structure.
2. The method for diagnosing a malfunction of a point switch machine as claimed in claim 1, wherein said sampling said current curve based on a preset length to obtain current curve data comprises:
under the condition that the length of the current curve is larger than the preset length, down-sampling the current curve to acquire current curve data;
and under the condition that the length of the current curve is smaller than the preset length, performing up-sampling interpolation on the current curve to obtain the current curve data.
3. The method of diagnosing a malfunction of a switch machine as claimed in claim 2, further comprising, before inputting the current curve data to a malfunction analysis network model:
acquiring a normal current curve sample and an abnormal current curve sample, and constructing a current curve sample set;
sampling each sample in the current curve sample set to obtain a current curve data sample set;
carrying out layered sampling on the current curve data sample set to obtain a current curve data training sample set;
enhancing the current curve data training sample set based on a sliding window method to obtain an enhanced current curve data training sample set;
combining the enhanced current curve data training sample set and the current curve data training sample set to form a model training sample set;
and pre-training the fault analysis network model by using the model training sample set and the diagnosis labels corresponding to the samples in the model training sample set.
4. The method for diagnosing faults of a switch machine as claimed in claim 3, wherein the enhancing the training sample set of current curve data based on a sliding window method to obtain the training sample set of enhanced current curve data comprises:
determining the size and the sliding step length of a sliding window;
sampling any current curve data in the current curve data training sample set based on the size of the sliding window and the sliding step length to obtain a plurality of enhanced samples to construct a local sample set;
the length of each of the enhancement samples in the local sample set is the size of the sliding window;
and (3) after each enhancement sample is supplemented with 0 backwards until the length of the enhancement sample is the preset length, acquiring the enhancement current curve data training sample set.
5. The method of diagnosing a malfunction of a switch machine as claimed in claim 3, further comprising, after pre-training the malfunction analysis network model:
verifying the pre-trained fault analysis network model by using a current curve data verification set;
the current curve data verification set is obtained after the current curve data sample set is subjected to layered sampling.
6. The method for diagnosing faults of a switch machine as claimed in claim 1, wherein the structure of the fault analysis network model includes:
the system comprises a total input layer, a first one-dimensional residual block, a maximum pooling layer, a one-dimensional Spatial Dropout layer, a second one-dimensional residual block, a global average pooling layer, a Softmax function layer and a total output layer which are connected in sequence;
the first one-dimensional residual block and the second one-dimensional residual block have the same structure.
7. The method of diagnosing a malfunction of a switch machine as claimed in claim 6, wherein the structure of said first one-dimensional residual block and said second one-dimensional residual block specifically comprises:
the multilayer optical amplifier comprises an interlayer input Layer, a first one-dimensional Layer Normailization Layer, a second one-dimensional Layer Normailization Layer, a third one-dimensional Layer Normailization Layer, an Add Layer Normailization Layer and an interlayer output Layer which are sequentially connected.
8. A switch machine fault diagnosis system, comprising:
the data acquisition unit is used for acquiring a current curve of the turnout switch machine;
the data preprocessing unit is used for sampling the current curve based on a preset length to obtain current curve data;
the data analysis unit is used for inputting the current curve data into a fault analysis network model so as to obtain a fault diagnosis result corresponding to the current curve according to an output result of the fault analysis network model;
wherein the fault analysis network model is a deep convolutional neural network with a residual error structure.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor when executing the program implements the steps of the method for diagnosing a malfunction of a switch machine as claimed in any one of claims 1 to 7.
10. A computer-readable storage medium, on which a computer program is stored, the computer program, when being executed by a processor, implementing the steps of the method for diagnosing a malfunction of a point switch according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011248062.6A CN112348170A (en) | 2020-11-10 | 2020-11-10 | Fault diagnosis method and system for turnout switch machine |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011248062.6A CN112348170A (en) | 2020-11-10 | 2020-11-10 | Fault diagnosis method and system for turnout switch machine |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112348170A true CN112348170A (en) | 2021-02-09 |
Family
ID=74363198
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011248062.6A Pending CN112348170A (en) | 2020-11-10 | 2020-11-10 | Fault diagnosis method and system for turnout switch machine |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112348170A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113627496A (en) * | 2021-07-27 | 2021-11-09 | 交控科技股份有限公司 | Method, device, electronic equipment and readable storage medium for predicting fault of turnout switch machine |
| CN113641486A (en) * | 2021-07-05 | 2021-11-12 | 西安理工大学 | Intelligent turnout fault diagnosis method based on edge computing network architecture |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105787511A (en) * | 2016-02-26 | 2016-07-20 | 清华大学 | Track switch fault diagnosis method and system based on support vector machine |
| CN106709567A (en) * | 2016-12-14 | 2017-05-24 | 河北省科学院应用数学研究所 | Method for diagnosing switch faults based on deep learning model |
| CN107451760A (en) * | 2017-09-04 | 2017-12-08 | 西安交通大学 | Based on when the limited Boltzmann machine of window sliding Fault Diagnosis of Roller Bearings |
| CN109711480A (en) * | 2018-12-30 | 2019-05-03 | 佳讯飞鸿（北京）智能科技研究院有限公司 | A kind of track switch gap monitoring device abnormal data method for detecting, apparatus and system |
| CN109902399A (en) * | 2019-03-01 | 2019-06-18 | 哈尔滨理工大学 | Rolling bearing fault recognition methods under a kind of variable working condition based on ATT-CNN |
| CN110728300A (en) * | 2019-09-09 | 2020-01-24 | 交控科技股份有限公司 | Method and system for identifying fault type based on turnout action current curve |
| CN111046583A (en) * | 2019-12-27 | 2020-04-21 | 中国铁道科学研究院集团有限公司通信信号研究所 | Switch machine fault diagnosis method based on DTW algorithm and ResNet network |
| CN111881950A (en) * | 2020-07-10 | 2020-11-03 | 交控科技股份有限公司 | Method and device for representing characteristics of current time sequence of turnout switch machine |
-
2020
- 2020-11-10 CN CN202011248062.6A patent/CN112348170A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105787511A (en) * | 2016-02-26 | 2016-07-20 | 清华大学 | Track switch fault diagnosis method and system based on support vector machine |
| CN106709567A (en) * | 2016-12-14 | 2017-05-24 | 河北省科学院应用数学研究所 | Method for diagnosing switch faults based on deep learning model |
| CN107451760A (en) * | 2017-09-04 | 2017-12-08 | 西安交通大学 | Based on when the limited Boltzmann machine of window sliding Fault Diagnosis of Roller Bearings |
| CN109711480A (en) * | 2018-12-30 | 2019-05-03 | 佳讯飞鸿（北京）智能科技研究院有限公司 | A kind of track switch gap monitoring device abnormal data method for detecting, apparatus and system |
| CN109902399A (en) * | 2019-03-01 | 2019-06-18 | 哈尔滨理工大学 | Rolling bearing fault recognition methods under a kind of variable working condition based on ATT-CNN |
| CN110728300A (en) * | 2019-09-09 | 2020-01-24 | 交控科技股份有限公司 | Method and system for identifying fault type based on turnout action current curve |
| CN111046583A (en) * | 2019-12-27 | 2020-04-21 | 中国铁道科学研究院集团有限公司通信信号研究所 | Switch machine fault diagnosis method based on DTW algorithm and ResNet network |
| CN111881950A (en) * | 2020-07-10 | 2020-11-03 | 交控科技股份有限公司 | Method and device for representing characteristics of current time sequence of turnout switch machine |
Non-Patent Citations (1)
| Title |
|---|
| 谷年龙: "基于残差网络的机械滚动轴承故障诊断方法研究", 《中国优秀硕士学位论文全文数据库工程科技Ⅱ辑》, no. 7, 15 July 2019 (2019-07-15), pages 2 - 1 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113641486A (en) * | 2021-07-05 | 2021-11-12 | 西安理工大学 | Intelligent turnout fault diagnosis method based on edge computing network architecture |
| CN113641486B (en) * | 2021-07-05 | 2024-03-01 | 西安理工大学 | Intelligent turnout fault diagnosis method based on edge computing network architecture |
| CN113627496A (en) * | 2021-07-27 | 2021-11-09 | 交控科技股份有限公司 | Method, device, electronic equipment and readable storage medium for predicting fault of turnout switch machine |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112034310A (en) | Partial discharge defect diagnosis method and system for combined electrical appliance | |
| CN111650453B (en) | Power equipment diagnosis method and system based on windowing characteristic Hilbert imaging | |
| CN112949715A (en) | SVM (support vector machine) -based rail transit fault diagnosis method | |
| CN112348170A (en) | Fault diagnosis method and system for turnout switch machine | |
| CN113591866B (en) | Special operation certificate detection method and system based on DB and CRNN | |
| CN110726898B (en) | Power distribution network fault type identification method | |
| CN115222650A (en) | Mixed industrial part defect detection algorithm | |
| CN116541790B (en) | New energy vehicle health assessment method and device based on multi-feature fusion | |
| CN112434566B (en) | Passenger flow statistics method and device, electronic equipment and storage medium | |
| CN112529109A (en) | Unsupervised multi-model-based anomaly detection method and system | |
| CN107392201A (en) | The pillar recognition methods of catenary mast, storage medium, processing equipment | |
| CN112215409B (en) | Rail transit station passenger flow prediction method and system | |
| CN113283550A (en) | Abnormal identification model training method for vehicle network electric coupling data | |
| CN114639102A (en) | Cell segmentation method and device based on key point and size regression | |
| CN113822336A (en) | Cloud hard disk fault prediction method, device and system and readable storage medium | |
| CN108613820A (en) | A kind of online allophone monitoring algorithm for GIS bulk mechanicals defect diagonsis and positioning | |
| CN116863122A (en) | Ammeter meter reading processing method, device, cloud end, system and medium | |
| CN116580232A (en) | Automatic image labeling method and system and electronic equipment | |
| CN113447572B (en) | Steel rail flaw detection method, electronic device, steel rail flaw detection vehicle and readable storage medium | |
| CN115601293A (en) | Object detection method and device, electronic equipment and readable storage medium | |
| CN114463300A (en) | Steel surface defect detection method, electronic device, and storage medium | |
| CN115115585A (en) | Cable fault diagnosis method and system and readable storage medium | |
| US20210089886A1 (en) | Method for processing data based on neural networks trained by different methods and device applying method | |
| CN113239075A (en) | Construction data self-checking method and system | |
| CN113658112B (en) | Bow net anomaly detection method based on template matching and neural network algorithm |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |